A Novel Form of the G Protein β Subunit Gβ5 Is Specifically Expressed in the Vertebrate Retina

The G protein β subunit, Gβ5, is predominantly expressed in the central nervous system. In rodent brain, Gβ5 is expressed as a protein with an apparent molecular mass of 39,000 daltons (39 kDa). We have identified an additional Gβ5 immunoreactive protein of apparent size 44 kDa in the vertebrate retina. Molecular cloning and sequencing of polymerase chain reaction products revealed that the cDNA encoding the larger species of Gβ5 (Gβ5L) was identical to the shorter form with the addition of 126 base pairs of 5′ DNA sequence potentially encoding an in-frame 42-amino acid extension. Sequencing of mouse Gβ5 genomic clones demonstrated that the 126-base pair of retinal-specific coding material is derived from a hitherto undetected 5′ exon. During sucrose density gradient fractionation of bovine retinas, the 44-kDa Gβ5L protein co-purified with rod outer segment membranes. Incubation of rod outer segment membranes with the nonhydrolyzable guanine nucleotide, GTPγS (guanosine 5′-3-O-(thio)triphosphate), which released the Gβ subunit of transducin (Gβ1), failed to remove Gβ5L. The 39-kDa Gβ5 protein displayed differential association with retinal and brain membranes. In the retina, Gβ5 was present as a soluble protein and was undetectable in the membrane fraction, whereas in the brain approximately 70% of Gβ5 was associated with cellular membranes. In transient COS-7 cell expression experiments, Gβ5L formed functional Gβγ dimers and Gαβγ heterotrimers, and activated phosphoinositide-specific phospholipase Cβ2 in a manner indistinguishable from the 39-kDa Gβ5 protein. The cloning of the retinal-specific Gβ5L cDNA suggests the existence of potentially novel G protein-mediated signaling cascades in photoreception.

In eukaryotic cells, a family of signal-transducing guaninenucleotide binding proteins (G proteins) 1 orchestrates many physiological processes by coupling activated cell surface recep-tors to intracellular second messenger systems. G protein-coupled receptors (GPCRs), which possess a stereotypical seventransmembrane-spanning domain architecture, bind and mediate the signaling of a variety of molecules, including hormones, neurotransmitters, odorants, and light. To date, several hundred GPCRs have been cloned or characterized (1). In contrast, the number of heterotrimeric G proteins, as well as the number of G protein-regulated effectors, is much more limited.
G proteins are heterotrimeric, composed of ␣, ␤, and ␥ subunits (G␣, G␤, G␥). Activation of a GPCR by ligand binding stimulates the exchange of bound GDP for GTP on the G␣ subunit and results in the dissociation of the G␣ subunit from a tightly complexed G␤␥ dimer. The released G␣ and G␤␥ subunits in turn regulate the activity of effector proteins, the better characterized of which include cGMP phosphodiesterase, adenylyl cyclases, phosphoinositide-specific phospholipase C ␤ enzymes, and ion channels. In addition, the dimeric G␤␥ subunits are involved in GPCR desensitization by recruitment of receptor kinases to the plasma membrane, and in signal transfer from seven transmembrane GPCRs to mitogen-activated protein kinase cascades (reviewed in Refs. 2 and 3).
The enormous diversity of the GPCR superfamily combined with the relatively limited set of G proteins and intracellular effectors poses problems for the generation of specific cellular responses to signaling molecules. It is believed that signal-sorting G proteins participate in the generation of highly specific and appropriate intracellular signals in several ways. Signaling of a particular receptor can be routed through specific G protein heterotrimers by somatic or temporal compartmentalization of the receptor and downstream signaling molecules with particular G protein subunits. For example, specificity in visual signal transduction by the light receptor, rhodopsin, is mediated in part by tissue-and cell type-specific expression of rod and cone cell G␣ and G␥ subunits (4). Somatic compartmentalization of other G protein ␣ and ␥ subunits has been described (5,6). To date, G␤ 5 , which is readily detected only in the central nervous system, is the only G protein ␤ subunit known to be expressed in a tissuerestricted fashion (7). Alternatively, GPCR signaling specificity may be achieved at the level of G protein-receptor coupling. There is a body of evidence to suggest that specific receptors display marked preferences for particular heterotrimeric combinations of G protein ␣, ␤, and ␥ subunits (8 -11). While it is known that not all G␤␥ dimeric combinations occur in vivo, there still remain a large number of combinatorial possibilities for G protein heterotrimer assembly. Thus, mechanisms to generate G protein subunit diversity have implications for signal transduction specificity.
Mammalian cDNA clones coding for at least 20 distinct G␣ subunit proteins have been described (12,13). Tandem duplication followed by evolutionary divergence of G␣ genes has been proposed as a fundamental genetic mechanism to increase the number and diversity of G␣ subunits (14). Alternative splicing has been shown to occur for certain G␣ gene transcripts and is an additional mechanism to increase G protein subunit diversity (15,16). In addition to the known G␣ subunits, cDNAs encoding 5 G␤ (7) and at least 12 G␥ (17) subunits have been described. The G␤ gene products are relatively similar, displaying amino acid identities ranging from 53 to 90%. The G␥ proteins, on the other hand, are much less well conserved.
In this paper, we report that the G␤ 5 gene is expressed as two distinct proteins in a tissue-specific manner. While G␤ 5 expression in brain is confined to a single 39,000-dalton (39-kDa) species, an additional G␤ 5 protein, with an apparent molecular mass of 44 kDa, is present in the vertebrate retina. We have cloned a retinal-specific G␤ 5 cDNA and show that this cDNA is identical to our previously described brain cDNA G␤ 5 clone with the exception of an additional 126 base pairs (bp) of sequence, encoding an additional 43 amino acids, at the 5Ј end. Sequence analysis of bacteriophage clones comprising part of the murine G␤ 5 genomic locus has revealed that the additional coding material derives from an upstream exon located approximately 2,900 bp (2.9 kilobase pairs) away from the initiator methionine of the 39-kDa brain G␤ 5 isoform. Fractionation of bovine retinas reveals that the long 44-kDa form of G␤ 5 (hereafter referred to as G␤ 5L ) is localized to the rod outer segment (ROS) membranes of the photoreceptor cell layer. While the function(s) of the retinally expressed G␤ 5 proteins is unclear, this unexpected versatility of the G␤ 5 gene has implications for G protein subunit diversity and the generation of complex, finely tuned signal transduction circuits.

EXPERIMENTAL PROCEDURES
Cloning, Sequencing, and Northern Analysis of Retinal G␤ 5 -Partial DNA sequence of a human retinal cDNA clone encoding a potentially longer form of G␤ 5 was initially identified during a search of The Institute for Genetic Research (TIGR) Human cDNA (HCD) data base using the full-length amino acid sequence of mouse G␤ 5 as the query. Using the DNA sequence of the TIGR clone as a guide, we cloned the coding region of the corresponding mouse cDNA by reverse transcription-polymerase chain reaction amplification (RT-PCR). Amplification of first strand cDNA prepared from mouse retina total RNA (a gift of J. Chen, Caltech) was carried out with Pfu DNA polymerase (Stratagene) using the following primers: 5Ј-ggaattcATGTNGATCAGACCTTTCT-3Ј and 5Ј-gctctagaTTATGCCCAAACTC-3Ј (N represents an equimolar ratio of all four bases, and lowercase letters are restriction enzyme recognition sites). The resulting amplification product was directionally cloned into the pcDNAI-amp expression vector (Invitrogen) and the entire sequence of several clones determined on both strands.
A 129SV mouse genomic DNA library in the cloning vector FIXII (Stratagene) was screened using the G␤ 5 cDNA as a probe according to standard methods (18). Isolated phage clones were rescreened with an oligonucleotide covering the first 18 bases of the G␤ 5 coding region to identify clones containing the 5Ј end of the gene. Positive clones were subcloned into pBluescript SKII (Stratagene), analyzed by restriction enzyme digestion, sequenced using dye-labeled primers on an Aplied Biosystems model 373 automated sequencer. One clone was identified that contained approximately 8 kb of 5Ј untranslated DNA. This 8-kb region was cloned into pBluescript SKII and partially sequenced by primer walking.
To isolate ROS, bovine retinas were fractionated as described (19) with minor modifications. Briefly, retinas were shaken on ice for 3 min in isotonic buffer (10 mM Tris-HCl, 100 mM KCl, 2 mM MgCl 2 , 1 mM DTT, pH 7.5) containing 45% sucrose. The suspension was then centrifuged at 5,000 ϫ g for 5 min. The floating ROS were collected by centrifugation at 15,000 ϫ g for 30 min after a 2-fold dilution of the sucrose with isotonic buffer. The crude ROS pellet was then subjected to ultracentrifugation on a stepwise sucrose density gradient, and the purified outer segments were recovered at the interface of the 1.115 and 1.135 density gradient steps. Protein samples (10 g) were separated on discontinuous SDS-polyacrylamide gels, and immobilized proteins were visualized as above using antiserum CT215 or the G␤ 1 -specific antiserum, BN1 (20).
Site-directed Mutagenesis and Transient Transfection-COS-7 cells were maintained in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) supplemented with 10% fetal calf serum at 37°C in an atmosphere containing 5% CO 2 . Expression vectors encoding PI-PLC␤ 2 , G␣ i2 , and various G protein ␤ and ␥ subunits used for transient transfection have been described previously (7,21). Expression vectors encoding mouse G␥ 4 (22) and rat G␥ 7 subunit (6) were the generous gifts of N. Gautam (Washington University, St. Louis, MO), and L. de Lecea and J. G. Sutcliffe (Scripps Institute, La Jolla, CA), respectively. The expression construct encoding mutant G␥ 2 ([Leu 71 ]G␥ 2 ) was provided by A. Katz. Codon 43 in G␤ 5L (methionine) was changed to encode alanine by site-directed mutagenesis with Pfu polymerase basically as described (23), and the resulting mutant, [Ala 43 ]G␤ 5L , was verified by DNA sequencing. Transient transfection of COS-7 cells using Lipo-fectAMINE (Life Technologies, Inc.) and inositol phosphate assays were performed as described previously (21). The level of inositol phosphates was determined 25 min after addition of 10 mM LiCl. For Western analysis, 1 ϫ 10 5 transiently transfected cells were scraped into 60 l of lysis buffer (50 mM HEPES, pH 7.4, 1 mM EGTA, 10 mM DTT, 0.2 TIU/ml aprotinin, and 0.5 mM phenylmethylsulfonyl fluoride), and membranes were pelleted at 15,000 ϫ g for 30 min in an Eppendorf centrifuge. Proteins were separated on 10% SDS-PAGE gels, transferred to nitrocellulose, and visualized as described above.

Identification and cDNA Cloning of a Retinal-specific Form of G␤ 5 -
We have reported previously that the G protein ␤ subunit, G␤ 5 , is expressed predominantly in the central nervous system. When immunoblots of crude membrane preparations from whole mouse brain, are probed with antiserum CT215, which recognizes a 16-amino acid peptide corresponding to the amino terminus of G␤ 5 , a protein of the appropriate molecular mass (39 kDa) can be readily detected. In contrast, G␤ 5 immunoreactive material is not detected on Western blots of membrane protein isolated from mouse heart, spleen, kidney liver, or skeletal muscle (7). 2 However, we had noticed that when crude membrane preparations from murine retinas were immunoblotted and probed with antiserum CT215, another immunoreactive protein, with an apparent molecular mass of 44 kDa was present, in addition to the 39-kDa G␤ 5 protein (Fig.  1). The presence of a larger protein immunologically related to G␤ 5 in retina can be explained in several ways. The larger protein could be transcribed from a gene closely related to G␤ 5 , it could result from post-translational modification of G␤ 5 , or, finally, it could be a splice variant of the G␤ 5 gene. Interestingly, the 44-kDa immunoreactive species was not observed in retinal protein samples obtained from rd mutant mice (Fig. 1, lane 3). Mouse strains carrying the rd mutation display retinal degeneration and a specific loss of rod photoreceptors (reviewed in Ref. 24). This result suggests that the 44-kDa protein, but not the 39-kDa G␤ 5 protein, is predominantly if not exclusively localized to rod cells in the retinal photoreceptor cell layer.
Concurrent with our identification of a larger protein immunologically related to G␤ 5 in retina, we had performed homology searches, using the amino acid sequence of mouse G␤ 5 to query the Institute for Genetic Research (TIGR) human cDNA expressed sequence tag (HCD EST) data base (25), as part of our on-going efforts to identify additional members of the G protein ␤ subunit family. This search returned 43 significant matches, 5 of which were determined to be partial sequences of the human G␤ 5 homologue. The DNA sequence of one of these (EST19480), which had been isolated from a human retinal cDNA library, was found to potentially encode a longer form of G␤ 5 . The DNA sequence of this partial cDNA clone overlapped bases 31-108 of G␤ 5 , which encode the first 36 amino acids of the protein. Significantly, EST19480 also had 193 bp of unrelated sequence at its 5Ј end, 126 bases of which had the potential to code for an additional in-frame 42 amino acids at the amino terminus of G␤ 5 .
Using the DNA sequence of EST19480 as a guide, we designed oligonucleotide primers to amplify the coding region of the corresponding mouse retinal transcript. Reverse transcription and amplification of total murine retinal RNA resulted in the appearance of the predicted 1,188-bp product, which was cloned and sequenced in its entirety. The sequence of this clone, which we have designated G␤ 5Long (G␤ 5L ), was identical to our previously reported mouse G␤ 5 cDNA sequence except for the addition of a 126-base pair extension at its 5Ј end. As determined for the EST19480 cDNA, the additional 126 bp of coding material in the G␤ 5L cDNA has the potential to direct the synthesis of a 42-amino acid peptide, which is in-frame with the rest of G␤ 5 sequence. The complete sequence of the G␤ 5L coding region is shown in Fig. 2. The G␤ 5L 1,185-bp open reading frame potentially encodes a protein of 395 amino acids with a predicted molecular mass of 43.6 kDa, in excellent agreement with the size of the retinal protein we had observed following SDS-PAGE. Homology searches of DNA and protein sequence data bases failed to identify any significant match to the retinal-specific coding material.
The retinal-specific 126-bp extension has no homology to the 5Ј-untranslated portion of G␤ 5 cDNAs isolated from mouse brain (7), 2 suggesting that it arises by retinal-specific usage of a 5Ј exon. We therefore isolated mouse genomic bacteriophage clones covering most of the G␤ 5 genomic locus and sequenced the first 3,923 bp upstream from the initiator ATG codon found in brain G␤ 5 transcripts. The beginning of the retinal-specific exon is located approximately 2,900 bp upstream of the first exon utilized in brain and is identical in sequence to the G␤ 5L RT-PCR product (save for two nucleotide alterations resulting from the upstream PCR primer sequence used for cloning) (Fig.  3). Consensus 5Ј and 3Ј splice donor and acceptor dinucleotide sequences are present at the beginning and end of the intron separating the first and second retinal exons (Fig. 3). In addition, the genomic sequencing revealed that the 5Ј-untranslated sequences present in brain G␤ 5 cDNAs are physically contiguous with the start of the 39-kDa G␤ 5 coding region (not shown). Sequence analysis of nearly 1 kilobase of genomic DNA sequence upstream of the G␤ 5L initiator ATG failed to reveal the presence of any known transcriptional control elements. However, the 2.8-kilobase pair intron between the first and second exons of G␤ 5L contains consensus sites for a variety of mammalian transcription factors and enhancer-binding proteins (not shown). We conclude that we have cloned the entire coding region of the mouse homologue of the partial human cDNA sequence obtained from the HCD EST data base, and conclude further that this transcript is produced by the addition of a single exon at the 5Ј end of the G␤ 5 coding sequence.
We previously reported that G␤ 5 mRNA is expressed in brain as two transcripts of approximately 1.75 and 2.3 kb (7). When total RNA isolated from mouse retina and mouse brain was probed with an end-labeled oligonucleotide corresponding to the first 43 bases of the G␤ 5L 5Ј-untranslated sequence, we observed hybridization only to retinal RNA (Fig. 4). The retinal G␤ 5L transcript is also present as two bands, with observed mobilities of 2.1 and 2.4 kb. The oligonucleotide probe did not hybridize to total brain RNA. Furthermore, when total brain RNA was analyzed by RT-PCR with primers corresponding to the 5Ј and 3Ј ends of the G␤ 5L coding region, no amplification of the 1,188-bp product was observed, even though this product was readily amplified from retinal RNA (data not shown). This result indicates that G␤ 5L is not expressed in brain and, coupled with our earlier results regarding the expression pattern of G␤ 5 , defines G␤ 5L as a retinal-specific gene product.
Retinal G␤ 5L , but Not G␤ 5 , Is Present in Rod Outer Segment Membranes-The observation that the 44-kDa G␤ 5L protein is specifically lost from the retinas of mice that have undergone retinal degeneration suggests that G␤ 5L is located in the photoreceptor cell layer. To characterize the retinal expression pattern of G␤ 5L further, we exploited the observation that ROS can easily be separated from the inner segments and soluble retinal proteins by means of centrifugation through sucrose density gradients (19). We used the polyclonal antiserum CT215 to follow the distribution of the two G␤ 5 proteins in these fractions and to compare it with the distribution of the ␤ subunit of transducin, G␤ 1 , which is known to be localized to the outer segment membrane. The results of these experiments are shown in Fig. 5. Upon low speed (5,000 ϫ g) centrifugation of gently homogenized bovine retina through 45% sucrose, the ROS membranes and soluble proteins remain in supernatant, while the inner segments and unbroken cells are found in the pellet. Both the short and long forms of G␤ 5 , as well as G␤ 1 , were found in this first supernatant fraction (S1). The ROS membranes can be separated from retinal soluble protein by dilution of the 45% sucrose supernatant and a second, 15,000 ϫ g centrifugation. At this point, the outer segment membranes are found in the pellet fraction and soluble proteins remain in the supernatant. When we further purified bovine ROS in this way, the short form of G␤ 5 was predominantly found in the soluble fraction (S2), while both the 44-kDa G␤ 5L and much of G␤ 1 were found in the pellet fraction (P2). G␤ 5L continued to co-purify with the ROS membranes and the membrane-associated G␤ 1 subunit after ultracentifugation on a stepwise sucrose density gradient (ROS) and could be only extracted from these membranes with detergents such as cholate, deoxycholate, and Lubrol (data not shown). When retinal membranes were washed more extensively, the 39-kDa G␤ 5 protein could not be detected in either the P2 or ROS fractions, even though the distribution of G␤ 1 was unchanged (data not shown).
It is known that incubation of ROS membranes with GTP promotes the release of the G␤ 1 ␥ 1 complex from the membrane fraction as a result of guanine nucleotide exchange on the transducin G␣ subunit and dissociation of the transducin heterotrimer. We tested whether guanine nucleotides had a similar influence on the membrane association of the 44-kDa G␤ 5L protein. However, as shown in Fig. 6, when we incubated ROS membranes with the nonhydrolyzable guanine nucleotide, GTP␥S, the 44-kDa G␤ 5 protein continued to remain completely associated with the membrane. As expected, incubation of the ROS membranes with GTP␥S effected the release of approximately 60% of G␤ 1 .
The 39-kDa G␤ 5 Brain Protein Is Membrane-associated-Our observation that the retinal 39-kDa G␤ 5 protein was essentially soluble and not associated with cell membranes was unexpected. We therefore examined the membrane association of the G␤ 5 protein in mouse brain, and compared its distribution to that of the G␤ 1 protein (Fig. 7). In contrast to our observations in retina, in brain both of the G protein ␤ subunits were associated with cell membranes. At least 70% of the detectable G␤ 5 protein in mouse brain was found in the membrane pellet (P), and the G␤ 5 present in this fraction could not be extracted in the absence of detergents (data not shown). A solution of 1% cholate, which solubilized about 50% of the total membrane proteins, was able to solubilize nearly all of G␤ 5 (compare supernatant and pellet fractions of the 1% cholate material in Fig. 7). As expected, all of the G␤ 1 protein in these samples was membrane-associated, and roughly 50% of this protein could be removed by cholate extraction.
Transient Expression of G␤ 5 Proteins in COS-7 Cells-The G␤ proteins are members of a much larger multiprotein family termed WD repeat proteins. The ␤ subunits of G proteins are able to form stable dimers with G protein ␥ subunits, and these G␤␥ dimers are capable of regulating the activity of several effector enzymes, including the ␤ 2 and ␤ 3 isotypes of phosphoinositide-specific phospholipase C. In contrast, non-G␤ WD repeat proteins do not dimerize with G␥ subunits nor do they activate G␤␥ effectors (26). To determine whether the G␤ 5L protein was capable of dimerizing with G␥ subunits, we transiently co-transfected COS-7 cells with cDNA expression plasmids encoding G␤ 5L and various G␥ subunits. The ability to form functional dimers was assessed by the ability of the coexpressed G␤ 5L and G␥ subunits to activate a co-transfected G␤␥ effector enzyme, PI-PLC␤ 2 . Our initial experiments were complicated by the fact that Western blot analysis of transfected cell extracts showed that COS cells expressed both short and long forms of G␤ 5 (Fig. 8A). This was apparently due to inappropriate initiation of translation at the second methionine codon (i.e. the initiator codon utilized in brain) in the transfected cells. To circumvent this problem, we altered the second methionine codon to encode alanine by site-directed mutagenesis. Western analysis of extracts prepared from COS-7 cells transfected with the mutant G␤ 5L ([Ala 43 ]G␤ 5L ) showed the presence of a single immunoreactive band that co-migrated with the 44-kDa retinal protein (Fig. 8A). We then asked whether [Ala 43 ]G␤ 5L could dimerize with various G␥ subunits and activate PI-PLC␤ 2 . The results of one such cotransfection experiment are shown in Fig. 8B. Co-transfection of either G␤ 5 or [Ala 43 ]G␤ 5L in combination with G␥ 2 resulted in a 4 -5-fold activation of PI-PLC␤ 2 . Co-transfection of other G␥ subunits with either of the G␤ 5 species resulted in less activation of the enzyme. In each case, the stimulation of PI-PLC␤ 2 -catalyzed inositol phosphate release was dependent on the presence of functional G␤␥ dimers. Replacement of the G␥ 2 protein with the [Leu 71 ]G␥ 2 mutant (Fig. 8B, G␥2*) in which the site of isoprenylation (Cys 71 ) was removed by site-directed mutagenesis reduced enzyme stimulation by approximately 60% (Fig. 8B). These results for activation of PI-PLC␤ 2 by the 39-kDa G␤ 5 are in agreement with our previous observations (7) and further demonstrate that the 44-kDa retinal G␤ 5L also possesses the ability to dimerize with G␥ proteins and activate a G␤␥ effector enzyme. An additional immunoreactive protein, with an apparent molecular mass of 32-35 kDa, was observed in extracts of COS-7 cells transfected with the G␤ 5L expression vectors (Fig. 8A). This may represent premature translational termination or carboxyl-terminal degradation of the G␤ 5L protein. However, since removal of COOH-terminal sequence from FIG. 6. Effect of GTP␥S upon elution of G␤ 5L from ROS membranes. Purified ROS membranes were washed once with isotonic buffer, twice with hypotonic buffer (10 mM Hepes-Na, pH 7.5, 0.5 mM EDTA, 1 mM DTT) and then with hypotonic buffer plus 100 M GTP␥S. The supernatants (S) and membrane pellets (P) from each step were analyzed by immunoblotting and detection with antisera specific for G␤ 5 (upper panel) or G␤ 1 (lower panel). The GTP␥S-treated supernatant is indicated by a "*" symbol. G␤ proteins results in a loss of G␤ function, 3 this smaller protein cannot account for the stimulatory activity observed in these experiments.
The presence of an extra 42 amino acids in the NH 2 -terminal part of the G␤ 5 subunit apparently does not affect dimerization with G␥ subunits or the activation of the downstream effector, PI-PLC␤ 2 . We next exploited the ability of a co-introduced G␣ i2 subunit to inhibit the activation of PI-PLC␤ 2 by scavenging free G␤␥ subunits (21) to determine whether we could detect differences in the abilities of the two G␤ 5 protein forms to interact with a G␣ subunit. The observed stimulation of PI-PLC␤ 2 by G␤ 5 ␥ 2 and [Ala 43 ]G␤ 5L ␥ 2 was decreased 71% and 59%, respectively, in the presence of G␣ i2 due to apparent suppression of signaling by the G␤␥ dimer as a result of heterotrimer formation (Fig. 8C). DISCUSSION The G␤␥ dimers of heterotrimeric G proteins have emerged in recent years as important regulators of ion channels, adenylyl cyclases, phosphoinositol-specific phospholipases, G protein-coupled receptor kinases, and mitogen-activated protein kinase cascades (2,3). Therefore, diversity within the G␤ subunit family has important implications for our understanding of cellular control of second messenger systems. With this in mind, we were interested to observe the presence of an additional protein immunologically related to the G protein ␤ subunit, G␤ 5 , in crude membrane preparations of mouse retina. To understand the molecular basis for this additional G␤ 5 -related protein, we have cloned, by RT-PCR, the cognate cDNA. Analysis of cDNA and genomic clones comprising part of the murine G␤ 5 locus revealed that the longer G␤ 5 isoform, G␤ 5L , results from retinal-specific utilization of a 5Ј exon located approximately 3 kb away from the initiator codon employed in brain G␤ 5 expression. During the sequencing of G␤ 5 cDNA and genomic clones, we detected two differences, at positions 516 and 534, from our previously published G␤ 5 sequence. A comparison of the sequencing data of G␤ 5 and G␤ 5L indicated that the sequence of the mouse G␤ 5L cDNA is in fact the correct one. Significantly, neither of the two missense "mutations" in our original sequence results in an amino acid change.
The G␤ 5L reading frame deduced from genomic sequencing is open for 192 bases upstream of our designated ATG initiator codon, and an in-frame methionine codon is present at position Ϫ189 (Fig. 3). Our assignment of the G␤ 5L initiator ATG is supported by the observation that the mobility of bona fide retinal G␤ 5L on SDS-polyacrylamide gels is the same as that obtained from our expression construct, [Ala 43 ]G␤ 5L (Fig. 8A). If the upstream methionine at Ϫ189 were used, we would expect that retinal G␤ 5L would be approximately 49 kDa, instead of the observed 44 kDa.
Several unanswered questions remain concerning the basis of expression of the multiple G␤ 5 isoforms. For example, the available data suggest, but we have not confirmed, that transcription of both G␤ 5 and G␤ 5L is directed by elements contained within the first intron, but that the two isoforms utilize different transcriptional start sites. It remains to be determined whether the expression of the 39-kDa G␤ 5 protein in retina is due to inappropriate initiation at the second methionine codon, as we observed in our COS-7 cell transfection experiments. Additionally, we still do not understand the basis for the two mRNA species seen on Northern analysis of retinal and brain RNA. Continued investigation into these phenomena may shed some light on mechanisms of retinal-specific gene expression.
In the retina, the 44-kDa G␤ 5L protein, but not the 39-kDa G␤ 5 protein, is associated with photoreceptor cell membranes. Because G␤ 5L is expressed only in retina and co-fractionates with rhodopsin-containing membranes, it is tempting to speculate that it is likely to be somehow involved in photoreception. However, the expression levels of G␤ 5 and G␤ 5L are much lower than that of the transducin G␤ 1 subunit (0.02-0.04% of total retinal and brain protein). 4 The fact that G␤ 5L cannot be eluted from the rod outer segment membrane by incubation with GTP␥S suggests that it may have a role there distinct from G␤ 1 . In this regard, it is interesting to recall that a fraction of 3  transducin ␣ subunit, which binds GTP␥S but does not dissociate from the rod membrane, has been described (27). Alternatively, it is possible that G␤ 5L is associated primarily with other G␤␥-binding proteins present in photoreceptors, such as phosducin, or is involved in regulation of downstream effectors, such as retinal PI-PLC (28).
The G␤ 5L protein is substantially larger than mammalian G␤ subunits identified previously. Indeed, the presence of an extended NH 2 -terminal domain provides a superficial resemblance to the yeast S. cerevisiae G␤ subunit, ste4 (29), and to certain members of the WD repeat multiprotein family (26), although this region of G␤ 5L has no homology to these proteins. Nevertheless, the results of our COS-7 cell co-transfection experiments clearly show that G␤ 5L is a bona fide G protein ␤ subunit and interacts with other members of the heterotrimeric G protein signal transduction cascade in a manner indistinguishable from G␤ 5 . This observed similarity in behavior is not unexpected since G␤ 5L and G␤ 5 share all residues involved in G␥ and G␣ contact (30,31). The fact that G␤ 5L and G␤ 5 display a marked preference for G␥ 2 in the PI-PLC␤ 2 activation assay is perhaps somewhat surprising in view of the recent report that G␤ 5 dimerizes rather poorly with G␥ 2 in a yeast two-hybrid assay (32). However, it should be remembered that our assay is a functional one and does not quantitate dimer formation.
An interesting question concerns the biochemical differences we observed for the 39-kDa G␤ 5 protein. Even though the 39-kDa G␤ 5 protein is expressed in both brain and retina, these proteins can be distinguished on the basis of their co-fractionation with cellular membranes, suggesting that G␤ 5 has different functions in these two compartments. In mouse brain, approximately 70% of G␤ 5 protein is associated with membrane fractions, whereas the retinally expressed 39-kDa G␤ 5 is found exclusively in the soluble cell fraction, even in the absence of detergents. G␤ proteins are targeted to the plasma membrane by virtue of their association with G␥ subunits. Dimerization of G␤ and G␥ proteins is followed by isoprenylation, proteolysis and carboxymethylation of the G␥ subunit, resulting in membrane localization of the G␤␥ dimer (reviewed in Ref. 33). It is thought that dimerization with a G␥ protein is essential for G␤ subunit stability and function. Some of the observed solubility of retinal G␤ 5 may be due to differences in the prenyl moiety associated with the gamma subunit since brain G␥ proteins are geranylgeranylated, while retinal G␥ 1 is farnesylated (33). However, this cannot completely account for the lack of membrane association of the 39-kDa G␤ 5 , since at least 50% of retinal G␤ 1 is found in the membrane pellet (Fig. 5). Perhaps retinal G␤ 5 is complexed with a non-isoprenylated G␥ protein.
A retinal endoproteolytic activity has been identified that is capable of producing a non-isoprenylated form of G␥ 1 (34). The difference in localization of G␤ 5 and G␤ 5L within the retina cannot be totally ascribed to differences in dimerization with G␥ proteins, however, since on the basis of the COS-7 cell co-transfection assay, both forms dimerize with G␥ proteins. Nevertheless, it seems reasonable to conclude that the 42amino acid NH 2 -terminal extension present in G␤ 5L is sufficient to allow efficient targeting of G␤ 5L to the outer segment membrane. This protein sequence may therefore prove to be a useful tool to study protein compartmentalization in the retina. Purification of native G␤ 5 and G␤ 5L proteins from retina should provide insight into the protein-protein interactions in which these subunits participate in vivo.
G␥ proteins reported to be present in bovine retina include G␥ 1 and G␥ 3 , which were found in rod and cone cell outer segments, respectively, and G␥ 7 , which was detected in the plexiform layers (4). In addition, a novel cone-specific G␥ subunit (G␥ 8 ) has been reported recently (35). Preliminary immunohistochemical analysis of mouse retinal sections with anti-serum CT215 indicates that G␤ 5 immunoreactivity is present in the plexiform layers, similar to the reported distribution for bovine G␥ 7 . 5 The absence of the G␤ 5L protein from the inner segment fraction of bovine retina (Fig. 5) suggests that this immunoreactivity may be due to the 39-kDa G␤ 5 protein, although other explanations, including species differences, are certainly possible. More detailed immunohistochemical and in situ hybridization experiments currently in progress should allow definitive sublocalization of the 39-and 44-kDa G␤ 5 proteins in retina and brain.